geek night 17.0 - artificial intelligence and machine learning

27
Artificial intelligence Machine learning

Upload: geeknighthyderabad

Post on 27-Jan-2017

109 views

Category:

Technology


0 download

TRANSCRIPT

Page 1: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Artificial intelligence Machine learning

Page 2: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Lee Sedol vs. AlphaGo

Silver, David, et al. "Mastering the game of Go with deep neural networks and tree search." Nature (2016)

Page 3: Geek Night 17.0 - Artificial Intelligence and Machine Learning

“At least a decade to go before a computer can beat a human expert”

Not very long AGo!

Page 4: Geek Night 17.0 - Artificial Intelligence and Machine Learning

1000 = 103

100000000000000000000000000000000000000000000000 = 1047

100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 = 10170

Why this assessment?

Page 5: Geek Night 17.0 - Artificial Intelligence and Machine Learning

What is AI?

https://xkcd.com/329/

Page 6: Geek Night 17.0 - Artificial Intelligence and Machine Learning

AI Winter

Series of setbacks from 70s till 90s

High Expectations

Failure of LISP machines

Failure of expert systems

Page 7: Geek Night 17.0 - Artificial Intelligence and Machine Learning

AI Spring

IMAGE: GETTY IMAGES/ISTOCKPHOTO

Page 8: Geek Night 17.0 - Artificial Intelligence and Machine Learning

AI Spring

Page 9: Geek Night 17.0 - Artificial Intelligence and Machine Learning

AI Spring

Domingos, Pedro. "A few useful things to know about machine learning." Communications of the ACM 55.10 (2012)

Page 10: Geek Night 17.0 - Artificial Intelligence and Machine Learning

What is ML?

“Field of study that gives computers the ability to learn without being explicitly

programmed”

Page 11: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Types of ML problems

Machine learning

Unsupervised learning

Supervised learning

Reinforcement learning

Regression Classification

Page 12: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Supervised Learning

Spoonfeeding labelled examples

Numerical values or Discrete class labels

Machine has to be ‘trained’ using a large corpus of ‘training data’

Page 13: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Regression

Training data

Hypothesis

Choosing optimum ‘hypothesis’ from training data

Hypothesis chosen has minimum ‘cost’

Typically used in financial applications, like predicting stock prices or likely monetary value of products

Page 14: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Classification

Height

Width

Decision boundary

Finding decision boundaries based on the labels of the training data

Non-linear decision boundaries require complex classifiers like SVMs and neural nets

Page 15: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Classification applications

Spam filtering

Optical Character Recognition

Pedestrian detection

Page 16: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Unsupervised Learning (Clustering)

Training data is not labelled

Grouping based on density (DBSCAN, OPTICS), cluster centers (K-Means) or probability distribution (GMM)

Page 17: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Clustering applications

Grouping similar news items Kharinov, M. "Hierarchical pixel clustering for image segmentation." arXiv preprint (2014).

Pixel clustering for segmentation

Page 18: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Reinforcement Learning

Teaching a machine by ‘rewarding’ it for good ‘actions’ and ‘punishing’ it for bad ones

Attempt is to explore the entire state space for a problem and get the best actions corresponding to each state, also known as ‘policy’

Page 19: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Reinforcement Learning applications

Reinforcement learning used for AlphaGo

Page 20: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Deep Learning

Capturing abstractions using a multi-level or ‘network’ approach

Each level or ‘layer’ composed of many simple processing units

The internal abstractions are often the best features to use for the problem, so no feature engineering is required

Page 21: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Artificial Neural Networks (ANNs)Deep networks composed of

artificial neurons

Inspired by biological neurons

Activation function is typically sigmoid, can be tanh or ReLu

The method used to train a network is called ‘backpropagation’

Traditional neural networks with all signals propagating in one direction are called ‘feedforward’ networks

Structure of a typical biological neuron

Typical artificial neuron

Page 22: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Artificial Neural Networks (ANNs) contd.

Rectifier functionLogistic function

Page 23: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Artificial Neural Networks (ANNs) contd.

Sigmoid function Typical feedforward neural network

Page 24: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Recurrent Neural Networks (RNNs)

Hidden layers feed back into themselves

Can be used to model sequences and for use as associative memory

Can take input sequences of arbitrary length using the concept of ‘attention’

Page 25: Geek Night 17.0 - Artificial Intelligence and Machine Learning

RNN applications (with links)Automatic music generation (Site has source code link)

Handwriting synthesis (Site has paper and source code links)

Intelligent personal assistants like Siri, Google Now, Cortana

Automatic image captioning

Sunspring

LSTM that generates poems

Page 26: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Learning Resources

Good courses or tutorials for ML

Coursera ML by Andrew NgDatacamp ML courseUdacity Deep Learning

Learning by doing

KaggleTopcoder Data science

Good video lectures for ML

Gilbert Strang lectures on Linear AlgebraNando de Freitas Deep Learning

Some people I follow in ML

Andrej Karpathy Peter NorvigAlex Graves Fei Fei LiAndrew Ng

Some good blogs on ML

WildMLIAmTraskKarpathy’s blog

And finally there’s Google Scholar. Read lots of

research papers and try to implement them!

Page 27: Geek Night 17.0 - Artificial Intelligence and Machine Learning

Thank YouHappy Learning :D